Form 5500 Headcount Estimates: A Reconciliation Guide

Executive Summary

Yes, Form 5500 can inform a private-company headcount estimate, but it cannot establish payroll headcount by itself. The useful fields are plan-participant observations with a defined plan, sponsor, reporting period, and coverage rule. They are not a universal employee count. On the 2025 Form 5500, line 6a(2) is the total active-participant count at plan-year end ^[1]. A pension-plan active participant is someone in covered employment earning or retaining credited service, including a worker eligible to elect 401(k) contributions even when the worker does not contribute ^[2].

The analytical task is therefore reconciliation, not copying. First define whether the mandate means consolidated employees, legal-entity employees, US employees, paid employees, full-time equivalents, or covered workers. Then resolve the sponsor's legal name, Employer Identification Number (EIN), plan number, entity group, geography, and plan type. Record the EIN and plan number together, but recognize that one participating employer may be named as sponsor for a controlled-group or multiple-employer plan ^[3]. Multiemployer plans are maintained by more than one employer ^[4]. Such filings rarely map cleanly to one acquisition target.

Time also matters. Form 5500 filings are normally due seven calendar months after plan-year end and an extension can add 2.5 months ^{[5] [6]}. Structured datasets are typically refreshed monthly and organized by form year, not necessarily plan year ^[7].

For a private-equity screen, the recommended output is a supported interval plus a decision. Pass only when the defensible lower bound clears the employee threshold. Fail only when the upper bound remains below it. Otherwise mark the company pending. Do not average conflicting definitions, and never sum multiple plans unless evidence shows that their populations are disjoint. This preserves what Docket describes as the source, supporting excerpt, and collection date for each researched answer (Source: docket.capital) while keeping unresolved questions visible rather than turning a participant count into false precision (Source: docket.capital).

Introduction and Background

Form 5500 is unusually valuable in private-company research because it is public, structured, recurring, and attached to an identifiable benefit plan. That strength often tempts analysts to treat an active-participant field as the answer to a different question: how many employees work for the company?

The substitution is unsafe. A plan count describes a covered population under plan rules. A company headcount describes a workforce under a chosen business definition. Those sets can overlap substantially without being identical. Eligibility waiting periods can omit recent hires. A plan may cover only a subsidiary, a union, a geography, or a class of workers. A controlled group may file through one sponsor. Former employees can remain in participant categories, while contractors and workers employed through a professional employer organization can sit outside the expected company or industry classification.

Even familiar official employment series use different units. Census Statistics of US Businesses (SUSB) counts full-time and part-time employees on payroll during the pay period containing March 12 ^[8]. The Bureau of Labor Statistics (BLS) Quarterly Census of Employment and Wages (QCEW) uses the pay period containing the 12th day of the month ^[9]. The Small Business Administration's employee-based size rules use pay periods across the preceding 24 months ^[10]. Each can be correct for its purpose and still yield a different number.

This report presents a field-level method for acquisition-origination teams applying an employee-size threshold or using headcount in a revenue model. It answers a narrow decision question: does the dated evidence support pass, fail, or pending? It does not turn a benefit filing into a diligence-grade employee census.

Define the Headcount Measure Before Searching

Six measures that analysts should not conflate

The first worksheet row should define the requested measure before any source is opened. At minimum, distinguish:

• Consolidated employees: workers across the parent and controlled subsidiaries intended by the investment mandate. Some federal rules generally treat commonly owned or related businesses as one employer ^[11].
• Legal-entity employees: workers employed by the exact sponsor or target entity, excluding sister entities unless the mandate includes them.
• US employees: workers whose work location falls within the United States, regardless of parent domicile.
• Paid employees: people on an employer payroll at the chosen reference date, commonly including full-time and part-time staff. QCEW excludes proprietors, unincorporated self-employed people, and unpaid family members ^[12].
• Full-time equivalents: a constructed workload measure, not a count of individual people. For one federal rule, non-full-time monthly hours are divided by 120 ^[13].
• Plan participants: people falling within a benefit plan's reporting definition and scope at a stated point in the plan year.

The difference between an establishment and an enterprise is equally important. Census defines an establishment as one physical location ^[14], while an enterprise may contain one or more domestic establishments under common ownership or control ^[15]. BLS also distinguishes an establishment from a firm or company ^[16]. The Bureau of Economic Analysis likewise cautions that an establishment is not necessarily the same as a company ^[17]. A location count, plan-sponsor count, and consolidated enterprise count can therefore refer to three different scopes.

Why employment definitions diverge

Definition choices affect whether a number can answer the mandate question:

• Part-time workers: SUSB paid employment includes full-time and part-time payroll workers ^[8]; a benefit plan may apply eligibility conditions.
• Contractors: SEC staff guidance recognizes that contractor-heavy businesses may need to distinguish retained independent contractors from regular employees ^[18].
• Common-law workers: IRS guidance says the substance of the relationship, not its label, governs worker status ^[19].
• Seasonal workers: a point-in-time observation can capture a peak or trough. One IRS seasonal exception uses 120 days or fewer, illustrating that seasonality can be rule-specific ^[20].
• Multiple jobs: Current Employment Statistics counts a person at each establishment where that person appears on payroll ^[21].
• Professional employer organizations: Census notes that workers may be classified outside a client business's predominant industry ^[22].

The practical definition should be written as a sentence, for example: "Total people on payroll across the target and controlled subsidiaries worldwide at the most recent quarter end, including part-time workers and excluding contractors." Every observation can then be classified against that sentence rather than treated as interchangeable.

Resolve the Employer, Plan, and Filing

Establish legal identity before using the count

Name matching alone is not enough. A search should begin with the legal entity, trade names, address, and known parent or subsidiaries. The Global Legal Entity Identifier Foundation (GLEIF) can provide an official name and registered address when a Legal Entity Identifier exists ^[23]. Its Level 2 data can identify direct and ultimate accounting-consolidating parents ^[24]. The data pool is available free of charge without registration ^[25], and its format records when information was most recently updated ^[26]. An LEI is an identity aid, not an employee source, and a lapsed record does not necessarily mean the entity is inactive ^[27].

Record these identifiers before interpreting a filing:

• Target legal name and any former or trade names.
• Target EIN, only when supported by a filing or authoritative document.
• Plan sponsor name and EIN exactly as filed.
• Plan number (PN), including leading zeros.
• Plan name and plan type, pension, welfare, or another arrangement.
• Entity relationship, exact target, parent, subsidiary, controlled group, multiple employer, or unresolved.
• Geography, especially when the mandate is US-only or global.

EIN continuity can help link filings, but it does not prove target-only coverage. A multi-unit Census enterprise may have several EINs ^[28].

Select the relevant filing, not merely the newest search result

For every candidate filing, record:

• Form year printed on the form.
• Plan-year beginning and ending dates.
• Date received by the filing system.
• Original, amended, short-year, or final status.
• Form type, Form 5500 or Form 5500-SF.
• EIN and plan number.
• Coverage scope and any uncertainty.

EFAST2 public search covers filings submitted since January 1, 2010 and does not require registration ^[29]. Search can use plan name, sponsor, EIN, plan number, acknowledgement ID, or combinations of criteria ^[30]. Public results show only the latest filing for each plan year ^[31].

An amended filing is a complete replacement, not a patch ^[32]. EFAST2 flags potential duplicates by matching identifiers including form year, EIN, and plan number ^[33]. In a bulk workflow, use the acknowledgement ID to join the form and schedules. Preserve prior versions for an audit trail, but use the latest valid replacement for the primary observation.

Exclude structures that do not map to one target

Set the filing aside or mark its scope unresolved when any of these applies:

• Multiemployer plan: participants span more than one contributing employer, so the plan total does not isolate the target ^[34].
• Defined contribution group arrangement: one consolidated filing can include a Schedule DCG for every participating plan ^[35].
• Controlled-group or multiple-employer sponsor: one participating employer may be listed as sponsor, which does not allocate participants to the target.
• Unclear welfare-plan structure: use governing documents and actual operations to determine whether benefits form one plan or several. The plan document defines who is covered ^[36].
• Missing public filing: some one-participant, foreign, sensitive, and older paper filings are outside public search coverage ^[37].

Interpret Participant Fields Without Calling Them Employees

Table 1 maps the principal 2025 fields to what they can and cannot support.

The table shows why active participants are the starting field, not the answer. Total participants and account-balance counts can be higher because they retain former workers. An active count can be lower than payroll if only some entities or employee classes are eligible. It can also be higher than the target's standalone headcount when a parent or controlled group sponsors the plan.

Pension and welfare definitions differ

For a pension plan that accepts employee contributions, voluntary or mandatory contribution can trigger participation ^[42]. Nonvested workers can qualify. Workers eligible to elect a 401(k) deferral can also be active participants even if they make no contribution. This makes the active field broader than "contributors" but still narrower or differently scoped than consolidated payroll.

For a welfare plan, making a contribution is one event that can establish participation ^[43]. That does not solve waiting periods, waived coverage, employer classes, or entity scope. The plan document determines who is covered ^[36].

Never sum plans without an overlap test

DOL explicitly warns that participant statistics can double-count workers who are in more than one plan ^[44]. A company's 401(k), pension, health, dental, and wrap arrangements may cover largely the same people. The form does not provide a cross-plan person identifier.

Apply this duplicate-plan check:

• Same sponsor and same broad employee class: presume material overlap.
• Pension plus welfare plans: do not add, even if values differ.
• Separate union and nonunion plans: add only if plan documents or company evidence show mutually exclusive groups.
• Separate subsidiaries: add only after confirming both belong in the mandate and populations are entity-specific.
• Different plan-year ends: do not add until dates are normalized.
• Unknown scope: retain both observations and mark the union unknown.

Reconcile Sources Into a Decision Range

The strongest output is a worksheet that keeps each observation intact. It should not average sources whose definitions differ.

Table 2 is the recommended reconciliation worksheet. The rows are generic examples of source types, not observations about a real company.

The worksheet forces the core questions before comparison: what, who, where, when, how stale, how authoritative, and whether populations overlap. Official statistical series help test plausibility but do not disclose a specific private company's workforce. The Census Business Register's individual-establishment data are not public ^[45]. QCEW covers more than 95% of US jobs, yet published output is principally an aggregate benchmark rather than a target record ^[46]. QCEW can also impute establishment data when reports are missing or poor quality ^[47] and uses proration for multiple-establishment employers ^[48].

Company and regulatory disclosures also need definition review. SEC human-capital disclosure is governed by materiality to understanding the business, rather than a universal employee-count specification ^[49]. A disclosed number may be authoritative for the issuer while remaining incomparable until its scope and date are read.

A repeatable reconciliation sequence

1. Normalize the requested measure. Write the employee definition, entity boundary, geography, and reference date.
2. Resolve identity. Match legal name, address, parent, EIN, and plan number. Preserve unresolved aliases.
3. Select the filing. Use the latest valid filing for the relevant plan year, while retaining amendment history.
4. Read the plan scope. Determine single-employer, multiple-employer, multiemployer, group arrangement, and covered classes.
5. Extract fields. Capture beginning and ending active counts, total participants, balance counts, and former-participant categories separately.
6. Calculate lag. Measure from plan-year end to the analyst's review date, not from website download date alone.
7. Add independent observations. Use dated company pages, filings, transaction materials, and professional-network ranges with their original labels.
8. Classify comparability. Mark each pair as same definition, convertible, directional only, or incomparable.
9. Test overlap. Never add plans unless evidence establishes disjoint populations.
10. Set bounds. Use the narrowest interval supported after definition and scope resolution, while retaining conflicts.
11. Apply threshold distance. Compare both bounds with the mandate threshold.
12. Return a status. Supported pass, supported fail, pending conflict, or unknown.

Confidence labels

Confidence should describe evidence fitness, not analyst conviction:

• High: exact entity and comparable employee definition, dated, direct, and independently corroborated.
• Medium: strong filing or company observation with one unresolved conversion, such as coverage ratio or geography.
• Low: indirect range, unclear entity scope, undated page, or self-associated profile count.
• Unknown: identity, scope, date, or population cannot be resolved sufficiently for the threshold decision.

Explicit validation states are useful, but they should not substitute for evidence. GAO recommends checking whether independent corroborating evidence supports conclusions drawn from data ^[50]. It advises more review when precise figures are demanded than when approximate data suffice ^[51].

Data Analysis and Evidence

Filing lag is measurable and material

A calendar-year plan ending December 31 can normally file at the end of the seventh following calendar month ^[5]. An authorized extension can move that to 9.5 months after year-end ^[6]. DOL's structured datasets then refresh around the first of each month. Consequently, an apparently current download can carry an active-participant observation that is well over a year older than the research date.

Other sources have different lags. QCEW is published about six months after its reference period ^[52], while SUSB can appear about 2 to 2.5 years after its reference year ^[53]. Census constructs enterprise employment by summing employment across associated establishments ^[54]. Those publications remain useful for market or industry context, but their lag disqualifies them as direct evidence of the target's current headcount.

Official sources also expose why apparently precise values need definition labels. SUSB excludes self-employed people and private-household employees ^[55]. SBA size rules count people employed full-time, part-time, or on another basis ^[56]. A separate IRS classification generally starts applicable-large-employer status at an average of 50 full-time employees, including full-time equivalents ^[57]. The definitions serve their own programs and should not be imported into an acquisition mandate without an explicit conversion.

Table 3 presents a fictional reconciliation and threshold-distance test. It is not based on any real company.

Assume a fictional mandate floor of 225 employees. The interval of 200 to 235 crosses the threshold. The threshold distances are 200 minus 225 = negative 25 at the low bound and 235 minus 225 = positive 10 at the high bound. The correct status is pending, not pass, fail, or an average of 217.5. The 17-person executive plan fails the duplicate check because its participants are a subset of the main population.

If the mandate floor were 175, the low bound would clear it by 25, supporting a provisional pass. If the floor were 250, the upper bound would remain 15 below it, supporting a provisional fail. These are screening statuses conditioned on the stated evidence, not assertions of exact payroll.

Why no accuracy percentage is justified

No public benchmark in this research validates Form 5500 active participants against a labeled sample of private-company payroll headcounts. Quoting a universal error rate would therefore be unsupported. A valid accuracy study would need:

• Ground truth: dated payroll headcount for the same entity, geography, and employee definition.
• Sampling design: companies spanning plan types, industries, sizes, and coverage arrangements.
• Version control: the exact filing and amendment used.
• Error decomposition: plan-coverage error, entity-scope error, date lag, and source-linkage error.
• Reported uncertainty: interval estimates and subgroup results, not only a pooled mean error.

NIST explains that a confidence interval provides lower and upper limits around an estimated mean ^[58] and that narrower intervals indicate greater precision ^[59]. A screening range here is not necessarily a statistical confidence interval. It is an evidence-bounded interval, and should be labeled accordingly.

Federal quality standards support the same discipline. Census identifies reference periods, reporting-unit definitions, and inconsistent administrative records as sources of measurement error ^[60]. It calls for key estimates to carry confidence intervals, margins of error, or equivalents ^[61], for data limitations to be stated ^[62], and for estimation methods to be disclosed ^[63]. Those are appropriate minimums for a headcount reconciliation record.

Implementation Guidance for Origination Teams

Store observations, not a single mutable headcount field

A research system should keep a current decision alongside an append-only observation table. Useful fields include:

• source_url and retained source copy;
• source_type and authority tier;
• observed_value_low and observed_value_high;
• measure_definition;
• entity_scope and entity_identifier;
• geography;
• as_of_date and collected_date;
• plan_year_begin and plan_year_end;
• form_year, received_date, and filing_status;
• ein, plan_number, and acknowledgement_id;
• confidence and overlap_risk;
• analyst_note and decision_status.

DOL's dataset guide says that each dataset adds system-generated filing-status, receipt-date, and signature-validity fields ^[64]. Pipelines should parse date formats explicitly and retain the original text.

Apply automated checks before analyst review

• Identity check: reject name-only matches when EIN, address, or parent conflicts.
• Version check: prefer the latest received valid replacement for the same plan year.
• Date check: flag observations older than the team's policy allows.
• Scope check: route multiple-employer, multiemployer, DCG, and unclear controlled-group filings to review.
• Overlap check: block arithmetic sums across plans by default.
• Threshold check: classify pass only when the low bound clears the floor and fail only when the high bound is below it.
• Evidence check: prevent a low-confidence range from producing an unconditional status.

This is consistent with Docket's stated model: identity and duplicate checks occur before detailed research, while a separate audit reviews conflicting sources and removes unsupported findings (Source: docket.capital) (Source: docket.capital). The point is reviewability, not a claim that automation resolves plan scope.

Implications and Future Directions

Form 5500 is best treated as a high-value observation layer. It is especially useful when a target has a stable single-employer plan, a clear sponsor relationship, recent comparable filings, and independent company evidence. It is weak when the target sits inside a complex group, workers are distributed across payroll entities, benefit eligibility is narrow, or the only accessible plan spans multiple employers.

Three improvements would materially strengthen future workflows:

• Coverage metadata: capture eligible classes, excluded entities, waiting periods, and geography when plan documents make them public.
• Longitudinal reconciliation: compare beginning and ending active counts across plan years, separating true workforce movement from amendments and plan changes. GLEIF's renewal process similarly calls for entity reference data to be revalidated annually ^[65].
• Validation sets: build consented, labeled samples in which payroll definitions and plan coverage are known, then report error by structure rather than one universal score.

The structured source itself will continue to change. Analysts should version field definitions, form instructions, and extraction logic rather than apply a present-day interpretation mechanically to older vintages.

More broadly, the best systems will preserve conflicting observations rather than overwrite them. OMB statistical standards emphasize giving users the principles and methods underlying a statistical product ^[66]. The same transparency belongs in acquisition screening: every range should show how it was built and what evidence could change the decision.

Frequently Asked Questions (FAQs)

Can Form 5500 estimate private-company headcount?

Yes, as one bounded input. An active-participant count can be a useful lower anchor or directional observation when the plan covers most employees and maps to the intended entity. It should not be copied into a CRM as "employees" without checking coverage, entity scope, geography, dates, and overlapping plans.

Which Form 5500 field is closest to employee count?

For either form, the total active participants at the end of the plan year field is usually closest. Total participants and participants with account balances are less suitable because they can include retired or separated people. The exact field labels appear in Table 1.

Why can active participants be lower than payroll headcount?

The plan may exclude subsidiaries, locations, worker classes, recent hires in a waiting period, or other ineligible workers. The IRS notes that the plan document describes who is covered. A company can also use multiple payroll entities or benefit arrangements.

Why can active participants be higher than the target's employees?

The sponsor may cover a controlled group or multiple employers, while the acquisition target is only one entity. A point-in-time target count may also differ from the filing's plan-year date. Separately, some plan fields include former workers, although those should not be confused with the active field.

How current is Form 5500 headcount data?

It can be materially delayed by the normal filing window, an authorized extension, and the next structured-dataset refresh. Always calculate age from the plan-year end represented by the participant field.

Should multiple plans be added together?

Usually not. DOL acknowledges double counting when workers participate in more than one plan. Sum only when evidence shows that populations are mutually exclusive and that every plan belongs within the requested entity and geography.

What is a reasonable confidence score?

Use categorical evidence labels instead of a made-up probability. High confidence requires a matching entity and definition, a known reference date, direct evidence, and independent corroboration. Medium confidence permits one material conversion. Low confidence applies when scope, date, or definition remains unclear. Unknown is appropriate when those questions prevent a threshold decision.

How should a PE team decide pass, fail, or pending?

Create a supported low and high bound. Pass when the low bound exceeds the mandate threshold. Fail when the high bound is below it. Pending when the range crosses the threshold or a material conflict remains. Keep a separate evidence-quality gate so a weak source cannot force a decision.

Conclusion

Form 5500 can improve a private-company headcount estimate, but only after the analyst translates a plan-level participant observation into the workforce definition required by the mandate. The decisive steps are to define the measure, resolve the legal entity and sponsor, select the right filing version, interpret active-participant fields in context, test multiple plans for overlap, and align every source by date, geography, and scope.

The method should end with a range and a research status. When the lower bound clears the employee floor, the evidence supports a provisional pass. When the upper bound remains below it, the evidence supports a provisional fail. When the range crosses the threshold or plan scope is unresolved, pending is the correct answer. No average can repair incompatible definitions.

For origination teams, the operational lesson is simple: retain the filed value, its exact field, the plan and entity identifiers, its plan-year date, the coverage interpretation, the independent observations, and the unresolved conflicts. Form 5500 is strongest as traceable evidence inside a reconciliation system. It is weakest when reduced to an unlabeled employee-count field.